Maboud F. Kaloorazi

Maboud F. Kaloorazi, Rodrigo C. de Lamare

An efficient, accurate and reliable approximation of a matrix by one of lower rank is a fundamental task in numerical linear algebra and signal processing applications. In this paper, we introduce a new matrix decomposition approach termed Subspace-Orbit Randomized singular value decomposition (SOR-SVD), which makes use of random sampling techniques to give an approximation to a low-rank matrix. Given a large and dense data matrix of size $m\times n$ with numerical rank $k$, where $k \ll \text{min} \{m,n\}$, the algorithm requires a few passes through data, and can be computed in $O(mnk)$ floating-point operations. Moreover, the SOR-SVD algorithm can utilize advanced computer architectures, and, as a result, it can be optimized for maximum efficiency. The SOR-SVD algorithm is simple, accurate, and provably correct, and outperforms previously reported techniques in terms of accuracy and efficiency. Our numerical experiments support these claims.

Maboud F. Kaloorazi, Rodrigo C. de Lamare

Low-rank matrix approximation plays an increasingly important role in signal and image processing applications. This paper presents a new rank-revealing decomposition method called randomized rank-revealing UZV decomposition (RRR-UZVD). RRR-UZVD is powered by randomization to approximate a low-rank input matrix. Given a large and dense matrix ${\bf A} \in \mathbb R^{m \times n}$ whose numerical rank is $k$, where $k$ is much smaller than $m$ and $n$, RRR-UZVD constructs an approximation $\hat{\bf A}$ such as $\hat{\bf A}={\bf UZV}^T$, where ${\bf U}$ and ${\bf V}$ have orthonormal columns, the leading-diagonal block of ${\bf Z}$ reveals the rank of $\bf A$, and its off-diagonal blocks have small $\ell_2$-norms. RRR-UZVD is simple, accurate, and only requires a few passes through $\bf A$ with an arithmetic cost of $O(mnk)$ floating-point operations. To demonstrate the effectiveness of the proposed method, we conduct experiments on synthetic data, as well as real data in applications of image reconstruction and robust principal component analysis.

Maboud F. Kaloorazi, Jie Chen

Matrices with low numerical rank are omnipresent in many signal processing and data analysis applications. The pivoted QLP (p-QLP) algorithm constructs a highly accurate approximation to an input low-rank matrix. However, it is computationally prohibitive for large matrices. In this paper, we introduce a new algorithm termed Projection-based Partial QLP (PbP-QLP) that efficiently approximates the p-QLP with high accuracy. Fundamental in our work is the exploitation of randomization and in contrast to the p-QLP, PbP-QLP does not use the pivoting strategy. As such, PbP-QLP can harness modern computer architectures, even better than competing randomized algorithms. The efficiency and effectiveness of our proposed PbP-QLP algorithm are investigated through various classes of synthetic and real-world data matrices.